---
title: Google NLP Salience: How Google Decides What Your Page Is About
description: How Google's salience scores define what a page is about, the four failure modes that dilute it, and how to engineer one entity per block so AI systems can extract and attribute claims correctly.
organization: Neural Command LLC
canonical: https://nrlc.ai/en-us/insights/google-nlp-salience/
---

← Back to insights

By Joel Maldonado, Semantic & Knowledge Architecture Researcher · SEO & AI visibility profile

Salience is computed from the text itself: how often an entity appears, where it appears, and how the surrounding language treats it.

An entity named in the title, defined in the opening, and carried through explicit references scores high. An entity mentioned once in passing scores low. The score is not a ranking factor you buy; it is a property of how the content is written.

A page with one strongly salient entity gives the machine a clear answer to "what is this about". That clarity feeds the classic results, knowledge features, and AI answers — extraction systems pull the salient claims, and attribution systems need a clearly identified subject to attribute them to.

- Entity dilution: ten topics in one article; no single entity rises above the rest, and the machine concludes the page is about nothing in particular.
- Pronoun dependence: "it", "this", "they" carry no entity signal; every pronoun is a missed attribution.
- Unattributed claims: a statement with no explicit subject cannot be cited to the entity that made it.
- Context-dependent paragraphs: a block that needs the block above it cannot be extracted, so it is dropped.
- One entity per content block; the block answers one question about that entity.
- Replace pronouns with explicit references at every opportunity.
- Write verbatim-quotable sentences: a claim extracted alone should still be complete and correct.
- Confirm the text with structured data: schema that names the same entities, with the same definitions.
- Measure extraction: if an AI system cannot restate the page's core claims with correct attribution, the salience engineering is not finished.
Neural Command's research treats citation as the primary metric: whether AI systems cite a page, how often, and whether the attribution is correct. Decision traces in generative search — reconstructing why a system cited what it cited — are part of that work, published on nrlc.ai.

Measurement of those citations lives in the AI search measurement cluster.

Salience is the input; correct citation is the output. Between them sits the entire discipline of AI visibility. If a page's text and structured data disagree about the entities, salience engineering cannot help — the free INGEST audit exposes exactly that gap.



---

Source: https://nrlc.ai/en-us/insights/google-nlp-salience/
Publisher: Neural Command LLC
License: Editorial use with attribution